Skip to content
ParamergeParamerge
Stephen Lieberman
Through Paramerge
Contact

PAN Lab example

Samagra Vedika

The match that cancels you: entity resolution as eligibility

A matcher links each person against thirty-plus government databases and decides eligibility from the merged profile, and a similarly-named stranger's car or land can be pulled onto you — silently flipping the flag that cancels your ration card, with no notice and no way to appeal. Modeled on Telangana's Samagra Vedika. The officials meant to catch the mistake defer to it as a backend 'technical' verdict. Watch where the leverage is: reconciling the match, not refining it.

Stylized model of a documented deploymentPublic benefits & eligibility

Open this example in PAN Lab v0.1 to apply pressures and levers and watch what the system does.

What this models

This example runs on the Samagra-Vedika-class entity-resolution eligibility matcher network: 6 components and 13 pathways between them. Every context in the Lab is a stylized model, never a reconstruction of any actual deployment, and each assumption behind it carries a provenance label.

Evidence base: 3 assumed · 4 published baseline. In the Lab, the shaded evidence band behind each headline readout draws its width from the least-established class below.

  • assumed

    This models the entity-resolution eligibility pattern documented in the Samagra Vedika case file — not a reconstruction of the actual matching engine or its undisclosed internals.

  • baseline

    The defining feature is an absence: nothing reconciled a matched asset against its true owner before the flag cancelled the card, so both the record-side reconciliation and the independent-second-match checks start closed at baseline. The documented error source was entity-resolution false-positive matching — a similarly-named third party's asset attributed to the applicant.

  • baseline

    The model-to-model self-loop encodes correlated error: one matching logic over roughly thirty million residents means a single fuzzy-match flaw (shared or common names colliding) repeats as correlated wrongful cancellations rather than averaging out.

  • baseline

    Officials were retained but deferring: the case documentation describes officials mandated to consult the matcher who declined to overturn it even against contrary evidence, so the operator correction pathways are present but weak (automation bias), not removed.

  • baseline

    The thirty-plus-linked-databases node carries a real inflow into the model — records joined across many government databases into the consolidated profile — so it enters the dynamics. What that linkage aggregated (records collected for other purposes) is documented in the case file, not computed here.

  • assumed

    The consolidated 360 profile is drawn as a mediating artifact on the match -> decision pathway, reflecting the case file's account of a single merged per-resident profile the eligibility flag was read off. It carries no flow of its own and does not affect the dynamics.

  • assumed

    The individual harms documented in the case file (a widow denied rations for over seven years after her deceased husband was linked to a third party's car; a family declared eligible only after a High Court ruling) and the harm's concentration on the poorest households are recorded in the case file, not computed here. This Lab models institutional workflow propagation, not demographics, and estimates no differential harm to served people.

What this example does not show

  • The individual harms and their concentration on the poorest households are documented in the case file. The Lab models institutional workflow propagation, not demographics, and estimates no differential harm to served people; that harm is documented in the case files and measured outside any diagram like this one.
  • The wrongful-rejection figure (roughly 7.5 percent, at least 15,471 of 205,734 re-processed cases) is a lower bound from an incomplete, court-ordered re-verification, not a full audit; the system's internals are proprietary and no accuracy data has been released, so the Lab uses the case's shape, not calibrated rates. The government's 95 percent 'efficiency' figure is a self-reported fraud-filtering measure, not a wrongful-exclusion rate.

Sources and evidence

What this example rests on, claim by claim. Every entry resolves to the same ledger the Evidence Registry publishes.

  • Samagra Vedika, an entity-resolution system built by the Telangana government, decided welfare eligibility by matching residents across thirty-plus government databases into a consolidated profile; between 2014 and 2019 more than 1.86 million ration cards were cancelled and 142,086 fresh applications were rejected without notice. Its core error was entity-resolution false-positive matching, in which a similarly-named third party's asset was attributed to the applicant and silently flipped the eligibility flag. After the Supreme Court of India ordered field re-verification in April 2022, a partial re-verification found roughly 7.5 percent wrongful rejection (at least 15,471 approved of 205,734 re-processed cases), a lower bound from an incomplete review; the system is proprietary and closed and an independent technical audit could not be completed, with no source code or accuracy data released. The government cited a self-reported 95 percent fraud-filtering efficiency, which measures spurious-application filtering rather than the wrongful-exclusion rate.

    empirical
    • Advocacy Amnesty International, Use of Entity Resolution in India: Shining a light on how new forms of automation can deny people access to welfare (2024) https://www.amnesty.org/en/latest/research/2024/04/entity-resolution-in-indias-welfare-digitalization/
    • Investigative Tapasya, Kumar Sambhav and Divij Joshi, How an algorithm denied food to thousands of poor in India's Telangana (Al Jazeera, with The Reporters' Collective and the Pulitzer Center AI Accountability Network) (2024) https://www.aljazeera.com/economy/2024/1/24/how-an-algorithm-denied-food-to-thousands-of-poor-in-indias-telangana
    • Academic Tushar V Sharma, Algorithmic Welfare Exclusion and the Right to Food in India: Lessons from Samagra Vedika (Oxford Human Rights Hub, University of Oxford) (2026) https://ohrh.law.ox.ac.uk/algorithmic-welfare-exclusion-and-the-right-to-food-in-india-lessons-from-samagra-vedika/
    • Trade press Sumit Jha, Telangana employs same tech to issue new ration cards that deleted 20 lakh names (The South First) (2024) https://thesouthfirst.com/telangana/telangana-employs-same-tech-to-issue-new-ration-cards-that-deleted-20-lakh-names/
    • Investigative Kumar Sambhav, Exclusive: Telangana offered its own 360 degree citizen tracking system to the Modi government (The Reporters' Collective; originally HuffPost India) (2020) https://www.reporters-collective.in/stories/exclusive-telangana-offered-its-own-360-degree-citizen-tracking-system-to-modi-govt
    • Investigative Pulitzer Center AI Accountability Network, How an algorithm denied food to thousands of poor in India's Telangana (2024) https://pulitzercenter.org/stories/how-algorithm-denied-food-thousands-poor-indias-telangana

  • A single automated rule set applied uniformly and without human review produced tens of thousands of correlated wrongful fraud determinations in the documented Michigan MiDAS case — one flaw repeating at caseload scale rather than averaging out.

    empirical
    • Government Michigan AG, settlement of civil-rights class action (Bauserman, 2022) https://www.michigan.gov/ag/news/press-releases/2022/10/20/som-settlement-of-civil-rights-class-action-alleging-false-accusations-of-unemployment-fraud
    • Investigative IEEE Spectrum, Michigan's MiDAS unemployment system: Algorithm alchemy that created lead, not gold https://spectrum.ieee.org/michigans-midas-unemployment-system-algorithm-alchemy-that-created-lead-not-gold

  • Under Samagra Vedika, exclusions were silent and there was no statutory route to contest an algorithmic decision, so the burden of proof fell on the excluded person: reporting describes officials who, though formally able to override the algorithm with evidence, deferred to it and declined to overturn its verdict, treating errors as backend technical issues. Documented individual harms include a 67-year-old widow denied rations for more than seven years after the system linked her deceased husband to a car owned by a similarly-named third person, and a family rejected for allegedly owning a four-wheeler that was declared eligible only after a Telangana High Court ruling. Corrections came through individual litigation and did not systematically feed back into the model, and the same entity-resolution technology was reused to issue new ration cards in 2024-2025.

    empirical
    • Investigative Tapasya, Kumar Sambhav and Divij Joshi, How an algorithm denied food to thousands of poor in India's Telangana (Al Jazeera, with The Reporters' Collective and the Pulitzer Center AI Accountability Network) (2024) https://www.aljazeera.com/economy/2024/1/24/how-an-algorithm-denied-food-to-thousands-of-poor-in-indias-telangana
    • Investigative The Reporters' Collective, A poor woman is declared rich; a living man dead. Their food and pension stopped by government (2024) https://www.reporters-collective.in/twitter-threads/a-poor-woman-is-declared-rich-a-living-man-dead-their-food-and-pension-stopped-by-government
    • Academic Tushar V Sharma, Algorithmic Welfare Exclusion and the Right to Food in India: Lessons from Samagra Vedika (Oxford Human Rights Hub, University of Oxford) (2026) https://ohrh.law.ox.ac.uk/algorithmic-welfare-exclusion-and-the-right-to-food-in-india-lessons-from-samagra-vedika/
    • Advocacy Amnesty International, Use of Entity Resolution in India: Shining a light on how new forms of automation can deny people access to welfare (2024) https://www.amnesty.org/en/latest/research/2024/04/entity-resolution-in-indias-welfare-digitalization/
    • Trade press Sumit Jha, Telangana employs same tech to issue new ration cards that deleted 20 lakh names (The South First) (2024) https://thesouthfirst.com/telangana/telangana-employs-same-tech-to-issue-new-ration-cards-that-deleted-20-lakh-names/

Where this connects

Institutional pressures in this domain

  • Austerity & recovery incentives — Cost-cutting and overpayment-recovery targets tilt the system toward denial and enforcement errors.
  • Vendor opacity — The deploying institution cannot inspect the model, data, or update pipeline it is accountable for.
  • Compliance over substance — Paper controls (sign-offs, checklists) satisfy audits while the behavior they describe erodes.
  • Reviewer bottleneck — One fixed-capacity checking stage sits between AI output and consequence; everything queues behind it.

All of them in context on the Public benefits & eligibility domain page.

Levers available here and the patterns behind them

Documented case histories